970 VS 980 (sli tri-sli)

[bestmak]

As we all know by now, the new NVIDIA 970 and 980 GPUs are the best thing to happen in a very long time.

I know that the 980 is better than the 970 in terms of RAW performance, but if one looks at price : performance ratio, the 970 wins here.

Alright, so now you say "I'm gonna buy a 970 today, but what if I want to sli in the future?" This is where you make some of the hardest decisions of your life. Yes, 970 sli is better than a single 980 (RAW POWER), but then 980 sli is better than 970 sli (no sh** sherlock) BUT then it costs more....

So I did a bit of research, and according to me, in terms of price:ratio : 970 > 980 & 970 sli > 980 & 970 sli > 980 sli...wait what? Well, I've noticed a very small (15 fps max?) FPS increase in 980 sli compared to 970 sli...for MORE THAN $450! You can by an extra 970 for that much...

ACTUAL QUESTIONS:

1) So that brings the question, in terms of RAW performance, is 970 tri-sli better than 980 sli since they would cost around the same?

2) Do you guys agree or disagree? Is 970 sli > 980 sli in terms of price:ratio AND is the 970 the winner here for single and sli in terms of price : performance?

 

[chenw]

1) If I were to choose between 970 tri-sli or 980 sli, strictly doing that comparison, I would go for 980 SLI. Reason being that the third card tend to not scale as well as the second card (often not nearly as well), and the hardware requirements for running a 970 tri sli is much higher than a 980 SLI. You will at least a more expensive motherboard, if not a more expensive platform, requires greater cooling management and a bigger PSU requirement. So in this case I would opt for 980 dual SLI.

2) Comparing 970 SLI and 980 SLI, I would agree. I can see the need for 980 SLI if one was going for 4k, but 970 SLI would be more ideal for 1440p 144hz resolution. the ~15% extra fps for the 980 is seldom going to make a game from unplayable to playable, so I would more than likely to save the money. Go 970's and opt for an earlier upgrade later down the line.

 

[Eggz]

I would skip out on both of the cards. They are GM204 chips, and Nvidia's chips ending in "4" tend to be immature versions they release to let consumers do testing. After a year, they apply feedback from reviews, surveys, and RMAs that they gathered from the "4" chips (which are locked down to provide less than full potential).

Then they release the "10" chip, which in this case would be the GM210. That will be the next generation of cards (maybe the GTX 1080 if they call it that). Then (sometimes) they release a "ti" version with a chip named "10B." The 780 ti was "GK110B" and reflected the best of what Kepler's architecture had to offer. Before that was the 780 (GK110 chip), and before that was the 770 and 680 (both a GK104 chip). Those were locked-down versions of Kepler. Nvidia is following the same pattern with Maxwell, so I'd just hold out unless your GPU is actually too slow to deal with.

If you can wait until the "GM210B," you'll get the best Maxwell with the most stable drivers they will make on that architecture.

 

[bestmak]

I would skip out on both of the cards. They are GM204 chips, and Nvidia's chips ending in "4" tend to be immature versions they release to let consumers do testing. After a year, they apply feedback from reviews, surveys, and RMAs that they gathered from the "4" chips (which are locked down to provide less than full potential).

Then they release the "10" chip, which in this case would be the GM210. That will be the next generation of cards (maybe the GTX 1080 if they call it that). Then (sometimes) they release a "ti" version with a chip named "10B." The 780 ti was "GK110B" and reflected the best of what Kepler's architecture had to offer. Before that was the 780 (GK110 chip), and before that was the 770 and 680 (both a GK104 chip). Those were locked-down versions of Kepler. Nvidia is following the same pattern with Maxwell, so I'd just hold out unless your GPU is actually too slow to deal with.

If you can wait until the "GM210B," you'll get the best Maxwell with the most stable drivers they will make on that architecture.

Thank you chenw and Eggz for your in depth answers. I have three small questions

I understand that these 2 cards just came out and we can predict future models of the graphics card but

1) How do we know that NVIDIA are "using" these cards for testing, then will release the next generation "proper" GPUs that use the chips full potential?

2) I heard people saying that NVIDIA will release a 8GB VRAM version of the 970 & 980, so will this happen for sure? Is it worth it to wait?

3) NVIDIA will always produce better cards year after year after year after year...we can always say that it's better to wait for the next gen chips/cards...but then we won't buy any graphics cards lol So my question really is, when should I buy a 970 or a 980 OR should I wait for the next gen cards (example: GTX 1080) if we are 100% sure that that will happen (and not after 5+years haha)?

 

[chenw]

1) Usually that is the cycle with nVidia, each new architecture will be first released as a lower tier card as a proof of concept (first Maxwell card was 750ti), then mainstream (currently 960, 970 and 980), then go enthusiast end with ti or Titan models. Lastly a new series with the Flagship card being the highest of that architecture. Fermi started out as 400 series and ended with 500, Kepler started with 600 and ended with 700's and titans.

2) It's currently speculation still, I wouldn't hold my breath for it. A higher VRAM card would make more sense on a more powerful card. There are plenty of reviews showing how little diff between 4GB and 8GB on 290x, both single and Xfired.

3) If you have the itch to buy now, 970 is a good buy. Any GPU will be outdated, just a matter of how soon, and TBH no one knows. 980 is just a poor buy IMO, I even expect 980ti to be better value than 980.

 

[Eggz]

(1) Like chenw mentioned, it's just the way chip manufacturers do things, and it's not just Nvidia. AMD and Intel do it also. Starting with the cheapest card for a new architecture makes sense. You don't want to charge $1,000 per chip until you've validated thousands of them. Nvidia starts with the mid-range cards (e.g. 750 ti for Maxwell), then enthusiasts for a couple generations (780 ti Kepler), and then business and professional cards (Quadro K6000 for Kepler). Maxwell is still too young for Nvidia to release as an unlocked chip to the prosumer and business community. By the time it matures to that level, Nvidia has worked out a lot of little kinks in the manufacturing process so that a percentage much closer to 100% of the chips will work without issue, and they're also better able to decipher which chips are "good" and which are just "acceptable" during the binning process (i.e. hand-picking the best chips from the lot). You see the same pattern with Intel. The Haswell chips came out first for cheaper electronics, then for the LGA1150 socket, then for the i7 on the LGA2011-3, and now for the Xeon on the LGA2011-3 socket. They just won't make contracts with 1,000 chips at $3,000 each unless they know for sure the things are going to be solid. Companies don't want to deal with returning chips, and they'll pay extra up front to not deal with it, but chip makers know that individuals like you and me don't mind tinkering and dealing with that stuff as much. We only have one or so computers to deal with, whereas companies can have thousands. That's especially if your stringing many chips together for a single supercomputer. The Dep't of Defense doesn't want to weeks figuring out which one of chips in their 750-chip machine is giving them trouble. They just want it to work. So chips on new architectures aren't available on that level until many, many every day consumers like you and me have "tested" them through purchasing for home use. That gives the chip manufacturers a chance to see where the bugs are on a level they can't produce in-house. How else can you test 100,000 chips to see what happens besides selling them? It sounds worse than it is, but that's part of the reason why i7 chips are cheaper than Xeon chips, even though they provide about the same level of computational power. It's also the reason why Quadro chips cost so much more than GeForce chips from Nvidia.

(2) Dito on what chenw said.

(3) Any released chip you buy from Nvidia's top of the GeForce lineup will be good. The 970 and 980 are both great cards on their own right. It's just that they don't represent the best of what Maxwell will offer for the reasons outlined in (1) above. I like to wait for the most mature cards in a lineup because it means your getting chips that are roughly as validated and stable as the Quadro cards, but they still cost considerably less. You benefit because you don't pay to test cards for Nvidia, and you also benefit from all of the testing that other did by virtue of buying less mature cards. As I said in my last post, though, get either the 970 or 980 if your current card needs an upgrade now.

 

[bestmak]

Fully understood both replies, thanks.

4) I would also want to buy the latest and best card in the lineup, but how would I know about it when it arrives? Is the best card in the series the Ti version?

5) Also, if I wait a pretty long time (I don't NEED an upgrade atm ) for the best GTX 900 series graphics card, would it be better to wait a bit more for the next gen cards (GTX 1080 for example)? Using information from previous series graphics cards by NVIDIA, when can we expect the best card in the line up AND the next gen graphics cards? 2 years? 5 years?

 

[Eggz]

(4) Two ways. First and foremost, look at the name of the chip. They are GX-xxx It's always G-something, for graphics. Kepler was GK for "Graphics Kepler." Maxwell is GM. The hundreds-place is the generation, meaning how many die shrinks the chip has undergone. So the 750 and 750 ti card were the first generation of Maxwell, so they had a chip called GM107. See link here: http://www.anandtech.com/show/7764/the-nvidia-geforce-gtx-750-ti-and-gtx-750-review-maxwell/4

Nvidia then shrunk the manufacturing process with the 970 and 980, so the hundreds place has a 2 rather than a 1 to denote that. So they have a chip called GM204. Like I mentioned in the previous post, a 4 in the ones place denotes a high-end but not fully unlocked chip. The chips with a 10 (e.g. GK110 in the 780 and Titan) represent the top chips, and IF Nvidia releases a ti version (which they don't always do), they will generally call its a 10B chip (e.g. GK110B in the 780 ti, Titan Black, and Quadro K6000). I've never seen them go past that before moving on to a new architecture.

So to answer your question, look for a card that has a GMx10 chip (might be the GM210, but if they shrink the die again for better efficiency, it might be called the GM310 . . who knows at this point besides insiders, right?). That brings up the second way to know - hardware reviews.

Legitimate hardware reviews like the one I linked above, and the ones that Toms Hardware releases will always tell you the name of the chip in a card. They will also tell you whether the chip is fully unlocked or not. Compare these from Tom's about the 780 and 780 ti (click underlined name of card to see full article):

SCHEMATIC VIEW OF THE GK110 CHIP IN THE GTX 780 (PARTIALLY DISABLED)

SCHEMATIC VIEW OF THE GK110B CHIP IN THE GTX 780 TI (FULLY UNLOCKED)

(5) They release flagship cards - those ending in 80 like the 580, 680, 780, and 980 - every year. Traditionally, the top GXx10 chips come out right after the GXx04 chips, so I'd expect the GM210 to be released a year from when they announced the 970 and 980.

 

[bestmak]

Ahh, I see.

6) So the top chips (GXx10) come out a year (educated guess) after the release of 970 and 980, but by around that time, the next series of GTX graphics cards will be released. Right?

7) Also do they release the cards ending in 70 every year too? Because I'm leaning towards the 970 because of its price : performance ratio. If they do release a 970 every year, in theory, I should just buy a GM210 970 when it is released, since it will be almost as good as the newly released next-gen graphics cards (like the 780 Ti compared to the 980)...unless they are better in terms of price : performance ratio (which is the case for the 970 compared to the 780 Ti right now)?

 

[Eggz]

(6) Yes'r!

(7) Yeah, the trends vary in exact timing and naming, so it might appear exactly as discussed, but you'll be able to recognize what's going when you see it. If they release a card that has a GM210 chip, my guess - and it is indeed a guess - would be for Nvidia to reserve that chip to the 80-version of the next card set. If they increase the hundreds place like they have been doing from year to year, then I'd guess Nvidia would reserve the new chip for the the GTX 1080. , not the 1070. That's just speculating, though.

 

[bestmak]

Thank you Eggz and chenw for helping me with some questions I had. I appreciate all the replies in this thread, and hope that if anyone else has the same question as I did, they can seek help from this thread.